• Teodor Dutu (26/26) Nov 08 2022 Hi,

Teodor Dutu <teodor.dutu gmail.com> writes:

Hi,

These weeks I finished implementing a template `_d_newitemU`. It 
creates an uninitialised `struct` object. It's used to lower 
expressions such as `new S(args)`, where `S` is a `struct`.

The compiler introduces the required `memset(p, 0, T.sizeof)` 
(for what was previously `_d_newitemT`) or `memcpy(p, T.init, 
T.sizeof)` (for what was previously `_d_newitemiT`) itself. This 
helps reduce the number of templates defined in DRuntime in the 
attempt to "unify" some instantiations of `_d_newitemU`.

I am now working to get the final tests for `new S(...)` working, 
**where `S` is a non-nested `struct`**. I haven't touched on 
nested structs yet as they require copying `vthis` and `vthis2`, 
which is cumbersome to do in `expressionsem.d`. Currently, these 
fields are set in `e2ir.d`, where there is [existing 
machinery](https://github.com/dlang/dmd/blob/master/compiler/src/d
d/toir.d#L444-L484) for duplicating them. I am now trying to "port" this code
to `expressionsem.d`, but I'm struggling with this because there is no 1:1
mapping between the backend and frontend constructions.

Regarding 
[`_d_arraycatnTX`](https://github.com/dlang/dmd/pull/14550), I 
feel the work is _nearly_ done. What's preventing buildkite to 
pass is that most likely there are cases in dmd where `.ptr` is 
used on strings that result from concatenations. These strings 
don't have a `\0` character at the end, hence [the seg 
fault](https://buildkite.com/dlang/dmd/builds/28559#0183fc10-644f-4cc9-a285-c9
b76ee122a/590-699). I can't reproduce the issue on my machine, so I'm kind of
left in the dark here.

Do you have any suggestions about how to find the cause of [this 
failure](https://buildkite.com/dlang/dmd/builds/28559#0183fc10-644f-4cc9-a285-c96b76ee122a)?

Thanks,
Teodor